Task-Dependent Algorithm Aversion
January 2020: We are compiling summaries of state-of-the-art research in ethics at the frontier of technology, following the theme of our 2019 Susilo Symposium. Today, we review insights on task-dependent algorithm aversion from Noah Castelo (University of Alberta), Maarten Bos (Snap Inc.) and Donald Lehmann (Columbia University).
In a previous blog post, we presented initial findings on algorithm aversion: after seeing an algorithm err, people preferred to rely on humans for forecasting MBA applicants’ future performance, even when doing so resulted in suboptimal forecasts. Subsequent research replicated algorithm aversion in other domains than applicants’ future performance, such as medical artificial intelligence, financial forecasts, employee selection or even joke recommendation.
In a new paper that we cover today, Castelo and colleagues argue that the willingness to use algorithms varies across different types of tasks, in particular if the task is more objective (involves facts that are quantifiable and measurable) or subjective in nature (is more open to interpretation and based on personal opinion or intuition).
Algorithm aversion is more pronounced for subjective (vs. objective) tasks
Castelo and colleagues identify a robust effect that algorithms are trusted and used less for tasks that are perceived as subjective in nature (e.g. dating) compared to tasks that are perceived as objective (e.g. financial advice). In one of the studies, the authors performed a field experiment using four Facebook ads, portraying either a human or an algorithm and providing either dating advice or financial advice. Participants were more likely to click on the dating advice ad when it was advertised as coming from a human versus an algorithm. However, there was no difference in clicks between human and algorithmic financial advice.
Framing a task as more objective increases willingness to use algorithms
Building on these results, Castelo and colleagues show that perceived task objectivity is malleable and can be the focus of practical marketing interventions. Specifically, describing a task as benefiting from quantitative analysis (rather than intuition) increases perceived task objectivity and consumers’ trust in and reliance on algorithms. In another study, the authors used two Facebook ads for an algorithm-based dating advice. Compared to the neutral ad just mentioning the algorithmic advice, the intervention ad highlighted the benefits of using a quantitative approach to dating advice (i.e., “studies show that using objective, quantifiable data is the best way to choose who to date”). The authors found that participants were more likely to click on the algorithmic advice dating ad framed as benefiting from a quantitative approach.
Overall, Castelo and colleagues propose practical interventions that marketers can use to increase consumers’ willingness to rely on algorithms instead of humans, by emphasizing the quantitative benefits of algorithmic advice.
The published academic paper can be found here:
Castelo, N., Bos, M. W., & Lehmann, D. R. (2019). Task-dependent algorithm aversion. Journal of Marketing Research, 56(5), 809-825.